Search Papers | Poster Sessions | All Posters
Poster C52 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Semantic decoding across participants and stimulus modalities
Jerry Tang1 (), Alexander G. Huth1; 1The University of Texas at Austin
Brain decoders that reconstruct language from semantic representations have the potential to improve communication for people with language disorders. However, training a semantic decoder for one participant currently requires many hours of brain data from that participant. In this study, we tested whether semantic decoders can be trained on data from separate reference participants, and then transferred to the goal participant. We functionally aligned the brains of the participants using responses to story and movie stimuli, and then applied the reference decoders to new responses from the goal participant. We found that cross-participant decoders outperformed within-participant decoders trained on the same amount of data. Notably, cross-participant decoding performance was high regardless of whether functional alignment was performed using story or movie stimuli. Our results demonstrate that cross-participant decoding can reduce the amount of training data required from the goal participant, and potentially enable decoding from participants who struggle to comprehend language stimuli.
Keywords: fMRI language decoding functional alignment